There’s a new section of the Stan web site, with case studies meant to illustrate statistical methodologies, classes of models, application areas, statistical computation, and Stan programming.
The first ten or so are up, including a grab bag of education models from Daniel Furr at U.C. Berkeley:
- Hierarchical Two-Parameter Logistic Item Response Model
- Generalized Rating Scale Model with Latent Regression
- Generalized Partial Credit Model with Latent Regression
- Rating Scale Model with Latent Regression
- Partial Credit Model with Latent Regression
- Rasch Model with Latent Regression
- Two-Parameter Logistic Item Response Model
There’s also an introduction to the R package that Daniel and crew (Seung Yeon Lee, Joon-Ho Lee, and Sophia Rabe-Hesketh) have been working on to encapsulate commonly used Stan models in education:
- Two-Parameter Logistic Item Response Model
and three from me:
- Pooling with Hierarchical Models for Repeated Binary Trials
- Multiple Species-Site Occupancy Model
- Soil Carbon Modeling with RStan
That’s not all
Meanwhile, Hiroki ITÔ has added Stan models up through Chapter 13 for the Kéry and Schaub book on population analysis.
To make this offer completely irresistible
I was inspired to finally get my act together after Matthew McKay and John Stachurski showed me QuantEcon’s Jupyter notebook gallery.
I introduced them to Jim Savage (not that one, this one), and they talked Jim into writing a notebook for them introducing Bayesian methods in general and Stan in particular. I’ve read the draft and it’s something I’m going to be recommending to a lot of people—I’ll announce here when it’s up.
Yours can be next
There are instructions on how to contribute a case study on the top-level Stan Case Studies page. They can be knitr (R), or Jupyter (R, Python, or Julia).
This is simply amazing stuff. Kudos to the stan team and the contributors to the case studies!
These are so useful, that you should add clear instruction how to cite these!
I was going back and forth on that. I didn’t want to give the impression that we were pretending to run a journal. But on second thought, I should really be treating this more like arXiv. So I’ll go back and add citation info. I think that’ll be more incentive for people to submit them, too.
I’d highly recommend writing them, too. It’s so nice to be able to do something self-contained and finite like this. knitr and Jupyter are both awesome for writing these little short-form, self-contained things.
We’ve been thinking about similar forms for the Stan book; an inspiring example of how to do it right is Hadley Wickham’s online R book, Advanced R, which is really going to be a pre-req for the Stan book, since I think we’ll almost certainly do it in R.
If you are having trouble getting Stan running in jupyter (I did), Matthew McKay put together a docker image that seems to work. You’ll need to install docker, then:
docker pull sanguineturtle/jupyter-rstan
docker run -d -p 8888:8888 sanguineturtle/jupyter-rstan
`docker-machine ip default` (will tell you the ip address)
and jupyter should be running on port 8888
(The current image has some of my notes baked into it; please disregard).
I’d cast a vote for org babel (http://orgmode.org/worg/org-contrib/babel/), too, in addition to knitr and Jupyter.